Methods and devices for demonstrating three-player pursuit-evasion (PE) game are provided using a hardware-in-loop test-bed. Robots including pursuer robots and an evader robot are arranged on a solid surface. A drone is positioned flying above to oversee the robots to capture a video or an image sequence of the robots. A robot thread process and a drone thread process are implemented by a computer. In the robot thread process, a tracking-by-detection process is perform to provide a state of the robot including a location and a heading direction of the robot; a delay compensation is conducted; and a PE game is called to calculate a robot command. In the drone thread process, a drone control is calculated to make the drone follow an evader robot, the drone control is sent to the drone, and user commands are also checked.
Legal claims defining the scope of protection. Each claim is shown in both the original legal language and a plain English translation.
1. A method for demonstrating a game theory by a hardware demonstrator, comprising: arranging robots, comprising pursuer robots and an evader robot, on a solid surface; positioning a drone flying above to oversee the robots to capture a video or an image sequence of the robots; in a robot thread process implemented by a computer and comprising a Timer thread process, on a first Timer sending a first command to a robot, obtaining an image of the robot after receiving the first command, the image being captured by the drone, performing a tracking-by-detection process to provide a state of the robot comprising a location and a heading direction of the robot, conducting a delay compensation, and calling a pursuit-evasion (PE) game to calculate a second robot command, wherein conducting the delay compensation comprises: determining delays between a time when an image is being captured by the drone and a time when the image is received by the computer, where the delays are measured using frames in a recorded video, and calculating a one-step-ahead state of the robot, based on the state of the robot and on the first calculated command to compensate the delays to provide delay-compensated state of the robot used when calling the PE game; and in a drone thread process implemented by a drone controller in the computer, capturing images of the robots, calculating locations of the robots from the images of the robots, calculating a drone control to make the drone follow an evader robot, sending the drone control to the drone, and checking user commands, wherein when an exit command is not issued, repeating process in the drone thread by first re-capturing images of the robots.
A method for demonstrating game theory using physical robots involves pursuer robots and an evader robot on a surface, overseen by a drone capturing video. A computer runs two processes. The robot process controls each robot using a timer, sending a command and then using drone imagery to track the robot's location and heading. It compensates for delays between image capture and processing by predicting the robot's one-step-ahead state based on current state and commanded action. A pursuit-evasion (PE) game algorithm then calculates the next command for the robot. The drone process captures robot images, determines their locations, and calculates drone movements to follow the evader, responding to user commands and repeating image capture if no exit command is given. The delay compensation involves measuring delays from video frame data.
2. The method according to claim 1 , wherein the tracking-by-detection process comprises: performing a background modeling to determine a background image, extracting regions of interest (ROIs) by a connected component algorithm, estimating an orientation of the robot in each ROI by a Histogram-based analysis of gradient distribution, detecting and classifying the robot in each ROI according to a best match score, wherein parameters with the best match score are used as detection result, and improving robustness by integrating temporal information and by integrating multiple target associations between target robots and robot templates.
The robot tracking process from the pursuit-evasion game demonstrator works as follows: First, it builds a model of the background. Then, it identifies regions of interest (ROIs) by identifying connected components in the foreground. Next, it estimates the orientation of robots in each ROI using histogram analysis of gradient distributions. The robots are then detected and classified in each ROI by template matching, selecting parameters based on the best match score. Finally, temporal information and multiple target associations between robots and templates are integrated to improve tracking robustness.
3. The method according to claim 2 , wherein performing the background modeling further comprises: subtracting the background image from each image frame in a recorded video to preserve pixels associated with the robot, and performing morphological operations to improve quality of background subtraction.
The background modeling step in the robot tracking system from the previous description subtracts the background image from each video frame, preserving robot pixels. It further refines the background subtraction through morphological operations which improves the quality and reduces noise in the resulting image.
4. The method according to claim 2 , wherein detecting and classifying the robot comprises: searching a best match in each ROI according to each robot template, wherein the searching comprises dynamic pruning, finding a robot template having the best match as the detection result, and fine-tuning for affine distortion comprising areas around the field boundary.
The robot detection and classification step of the robot tracking system, as described previously, searches for the best match for each robot template within each region of interest (ROI). This search utilizes dynamic pruning techniques to improve efficiency. The robot template with the best match is chosen as the detection result. To improve accuracy, fine-tuning is performed to account for affine distortions, particularly in areas near the playing field boundary.
5. The method according to claim 1 , further comprising a learning method to estimate parameters by: (a) recording an initial state (x 0 , y 0 , h 0 ) of the robot using the tracking-by-detection process from images captured by the drone, (b) sending a command to the robot, (c) obtaining a first location and heading state (x 1 , y 1 , h 1 ) of the robot by tracking algorithms, (d) calculating a first state change (Δx 1 , Δy 1 , Δh 1 ), where Δx 1 =x 1 −x 0 , Δy 1 =y 1 −y 0 , and Δh 1 =h 1 −h 0 , (e) sending an opposite command to the robot, (f) obtaining a second location and heading state (x 2 , y 2 , h 2 ) of the robot after the opposite command by the tracking algorithms, (g) calculating a second state change (Δx 2 , Δy 2 , Δh 2 ), where Δx 2 =x 2 −x 1 , Δy 2 =y 2 −y 1 , and Δh 2 =h 2 −h 1 , and (h) repeating steps (a)-(g) for a plurality of times to provide a plurality of the first state changes and a plurality of the second state changes, and computing an average of each of the first state changes (Δx 1 , Δy 1 , Δh 1 ) and the second state changes (Δx 2 , Δy 2 , Δh 2 ).
To automatically estimate parameters for accurate robot control, the system performs a learning method. It starts by recording the robot's initial location and heading. A command is sent to the robot. The robot's new location and heading are recorded. The difference between the initial and new states is calculated. An opposite command is sent, and the resulting location and heading are measured again. The difference between the states after the first and second commands is calculated. This process is repeated multiple times to collect many state change measurements, and the average state changes for each command direction are computed.
6. The method according to claim 1 , wherein calculating the locations of the robots from the images of the robots in the drone thread process does not calculate the heading of the robots to save a large amount of computing time.
To reduce computation time in the drone's image processing, the system described earlier only calculates the location of the robots, and omits the heading calculation, when calculating robot positions from drone images within the drone thread process.
7. The method according to claim 1 , wherein calling the PE game to calculate the second robot command in the robot thread process comprises: sending the states of the robots by performing the tracking-by-detection process to a three-player PE game model to solve the states of the robots comprising the two pursuer robots and one evader robot using a game equilibrium in a game solution, wherein the game equilibrium provides the second robot command respectively to the pursuer robots and the evader robot.
The pursuit-evasion game algorithm used in the robot control system described previously receives robot states (location and heading) as input. This state data is obtained from the tracking-by-detection process. The game model solves for the optimal strategies of the pursuer and evader robots, using a game equilibrium solution, which then provides the next command for each robot.
8. The method according to claim 1 , further comprising: sending a plurality of scenario configurations to a pursuer agent and an evader agent running on the computer to test the game theory.
The system for demonstrating game theory as described previously allows for testing various scenarios by sending different scenario configurations to the pursuer and evader agents running on the computer.
9. A method for demonstrating a game theory by a hardware demonstrator, comprising: arranging robots, comprising pursuer robots and an evader robot, on a solid surface; positioning a drone flying above to oversee the robots to capture a video or an image sequence of the robots; in a robot thread process implemented by a computer and comprising a Timer thread process, on a first Timer sending a first command to a robot, obtaining an image of the robot after receiving the first command, the image being captured by the drone, performing a tracking-by-detection process to provide a state of the robot comprising a location and a heading direction of the robot, conducting a delay compensation, and calling a pursuit-evasion (PE) game to calculate a second robot command; in a drone thread process implemented by a drone controller in the computer, capturing images of the robots, calculating locations of the robots from the images of the robots, calculating a drone control to make the drone follow an evader robot, sending the drone control to the drone, and checking user commands, wherein when an exit command is not issued, repeating process in the drone thread by first re-capturing images of the robots; and setting a timer for robot controls in the robot thread process, wherein an execution duration is set for the robot controls to provide sufficient time for performing the tracking-by-detection process and for conducting the delay compensation.
A method for demonstrating game theory uses physical robots: pursuer robots and an evader robot on a surface, overseen by a drone capturing video. A computer controls each robot using a timer, sending a command and then using drone imagery to track the robot's location and heading. It compensates for delays between image capture and processing. A pursuit-evasion game calculates the next command. The drone process captures robot images, determines their locations, and calculates drone movements to follow the evader. A timer sets the duration for robot controls in the robot thread process, allowing adequate time for tracking and delay compensation.
10. The method according to claim 9 , wherein the tracking-by-detection process comprises: performing a background modeling to determine a background image, extracting regions of interest (ROIs) by a connected component algorithm, estimating an orientation of the robot in each ROI by a Histogram-based analysis of gradient distribution, detecting and classifying the robot in each ROI according to a best match score, wherein parameters with the best match score are used as detection result, and improving robustness by integrating temporal information and by integrating multiple target associations between target robots and robot templates.
The robot tracking process from the pursuit-evasion game demonstrator (as described in claim 9) works as follows: First, it builds a model of the background. Then, it identifies regions of interest (ROIs) by identifying connected components in the foreground. Next, it estimates the orientation of robots in each ROI using histogram analysis of gradient distributions. The robots are then detected and classified in each ROI by template matching, selecting parameters based on the best match score. Finally, temporal information and multiple target associations between robots and templates are integrated to improve tracking robustness.
11. The method according to claim 10 , wherein performing the background modeling further comprises: subtracting the background image from each image frame in a recorded video to preserve pixels associated with the robot, and performing morphological operations to improve quality of background subtraction.
The background modeling step in the robot tracking system from the previous description (as described in claim 10) subtracts the background image from each video frame, preserving robot pixels. It further refines the background subtraction through morphological operations which improves the quality and reduces noise in the resulting image.
12. The method according to claim 10 , wherein detecting and classifying the robot comprises: searching a best match in each ROI according to each robot template, wherein the searching comprises dynamic pruning, finding a robot template having the best match as the detection result, and fine-tuning for affine distortion comprising areas around the field boundary.
The robot detection and classification step of the robot tracking system, as described previously (as described in claim 10), searches for the best match for each robot template within each region of interest (ROI). This search utilizes dynamic pruning techniques to improve efficiency. The robot template with the best match is chosen as the detection result. To improve accuracy, fine-tuning is performed to account for affine distortions, particularly in areas near the playing field boundary.
13. The method according to claim 9 , further comprising a learning method to estimate parameters by: (a) recording an initial state (x 0 , y 0 , h 0 ) of the robot using the tracking-by-detection process from images captured by the drone, (b) sending a command to the robot, (c) obtaining a first location and heading state (x 1 , y 1 , h 1 ) of the robot by tracking algorithms, (d) calculating a first state change (Δx 1 , Δy 1 , Δh 1 ), where Δx 1 =x 1 −x 0 , Δy 1 =y 1 −y 0 , and Δh 1 =h 1 −h 0 , (e) sending an opposite command to the robot, (f) obtaining a second location and heading state (x 2 , y 2 , h 2 ) of the robot after the opposite command by the tracking algorithms, (g) calculating a second state change (Δx 2 , Δy 2 , Δh 2 ), where Δx 2 =x 2 −x 1 , Δy 2 =y 2 −y 1 , and Δh 2 =h 2 −h 1 , and (h) repeating steps (a)-(g) for a plurality of times to provide a plurality of the first state changes and a plurality of the second state changes, and computing an average of each of the first state changes (Δx 1 , Δy 1 , Δh 1 ) and the second state changes (Δx 2 , Δy 2 , Δh 2 ).
To automatically estimate parameters for accurate robot control (as described in claim 9), the system performs a learning method. It starts by recording the robot's initial location and heading. A command is sent to the robot. The robot's new location and heading are recorded. The difference between the initial and new states is calculated. An opposite command is sent, and the resulting location and heading are measured again. The difference between the states after the first and second commands is calculated. This process is repeated multiple times to collect many state change measurements, and the average state changes for each command direction are computed.
14. The method according to claim 9 , wherein calculating the locations of the robots from the images of the robots in the drone thread process does not calculate the heading of the robots to save a large amount of computing time.
To reduce computation time in the drone's image processing, the system described earlier (as described in claim 9) only calculates the location of the robots, and omits the heading calculation, when calculating robot positions from drone images within the drone thread process.
15. A hardware demonstrator device for demonstrating a game theory, comprising: robots, comprising pursuer robots and an evader robot, placed on a solid surface; a drone, flying above to oversee the robots to capture a video or an image sequence of the robots; and a computer configured with a pursuer agent and an evader agent and configured to perform a Timer thread process to: send a first command to a robot, obtain an image of the robot after receiving the first command, the image being captured by the drone, perform a tracking-by-detection process to provide a state of the robot comprising a location and a heading direction of the robot, conduct a delay compensation, and call a PE game to calculate a second robot command; and a parameter estimator, configured to compute an average of a first state change (Δx 1 , Δy 1 Δh 1 ) based on a command to the robot over an initial sate, and a second state change (Δx 2 , Δy 2 , Δh 2 ) based on an opposite command to the robot and the command, wherein a drone controller operated on the computer is configured to: control the drone to capture images of the robots, calculate locations of the robots from the images of the robots, calculate a drone control to make the drone follow an evader robot, send the drone control to the drone, and check user commands.
A hardware demonstrator for game theory includes pursuer and evader robots on a surface, and a drone overhead capturing video. A computer is configured with pursuer and evader agent software. The computer performs a timer-based process: sends a command to a robot, obtains a drone image of the robot, tracks the robot's location and heading, compensates for delays in image processing, and runs a pursuit-evasion game to calculate the next command. A parameter estimator averages state changes from robot commands and opposite commands. A drone controller calculates drone movements to follow the evader, and checks user commands.
16. The device according to claim 15 , wherein each robot comprises a wireless radio, built in drive commands, and a sensor including a laser and a camera.
Each robot within the hardware demonstrator described previously includes a wireless radio for communication, built-in drive commands for movement, and a sensor suite comprising a laser and a camera for environmental perception.
17. The device according to claim 15 , wherein each robot is configured for sending the image or the video to the computer, and for commanding acknowledge and response to a corresponding agent on the computer.
Each robot in the hardware demonstrator described previously is configured to wirelessly transmit images or video to the central computer, and to send acknowledgement and response signals to corresponding agent software running on the computer.
18. The device according to claim 15 , wherein the computer is configured to send commands for moving, commands for camera, and commands for laser to each robot.
The computer in the game theory demonstrator described previously is configured to send commands to each robot, which may include movement commands (e.g., speed, direction), camera commands (e.g., capture image), and laser commands (e.g., activate laser).
19. The device according to claim 15 , wherein the drone-controller is a proportional-integral-derivative (PID)-based drone controller.
The drone controller within the hardware demonstrator described previously uses a proportional-integral-derivative (PID) control algorithm to regulate the drone's movement and keep the evader robot within the drone's field of view.
20. The device according to claim 15 , further comprising a graphical user interface (GUI) with a scenario manager on the computer.
The hardware demonstrator described previously also includes a graphical user interface (GUI) with a scenario manager that allows users to define and load different game scenarios or configurations to the computer.
Cooperative Patent Classification codes for this invention. Click any code to explore related patents in that topic.
October 31, 2014
August 22, 2017
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.